{
  "generated_at": "2026-10-01T00:00:00Z",
  "status": "preview installers available; exact release model qualification pending",
  "current_release_tiers": {
    "laro_lite": {
      "checkpoint": "unsloth/gemma-3-4b-it-GGUF · Q4_K_M",
      "sha256": "04a43a22e8d2003deda5acc262f68ec1005fa76c735a9962a8c77042a74a7d19",
      "bytes": 2489894016,
      "status": "historical MLX run exists; equivalence to this GGUF/runtime has not been established"
    },
    "laro_med": {
      "checkpoint": "unsloth/gemma-3-12b-it-GGUF · Q4_K_M",
      "sha256": "15b8fd9d8672cd4240c178c217ca781409291f34e353d2e913b29c7602ceb3ff",
      "bytes": 7300778336,
      "status": "no complete release-attributed standardized score"
    },
    "laro_max": {
      "checkpoint": "unsloth/gemma-3-27b-it-GGUF · Q4_K_M",
      "sha256": "f1b699659942c777bd3ec0bcb527d6ebf34ae14ca76e3af103d58d0c9cbdadee",
      "bytes": 16546688736,
      "status": "no complete release-attributed standardized score; 24 GB minimum, 32 GB recommended"
    }
  },
  "historical_humaneval": {
    "protocol": "HumanEval 164 tasks; one generated completion per task; generated code executed; incomplete output counted as failure",
    "gemma_3_4b_base": {
      "result": "113/164",
      "score_pct": 68.9,
      "runtime_identity": "mlx-community/gemma-3-text-4b-it-4bit (base)",
      "report": "/evidence/historical/humaneval-gemma3-4b-base.json",
      "report_sha256": "40e24408fe3895b7bb50026c86e4d6e18d1beda752257d286beb56488dbb97f7",
      "inherited_by_current_laro": false
    },
    "gemma_4_12b_ollama": {
      "result": "149/164",
      "score_pct": 90.9,
      "runtime_identity": "gemma4:12b via Ollama",
      "report": "/evidence/historical/humaneval-gemma4-12b-ollama.json",
      "report_sha256": "92451589b7f897005d8bdada66c76fea836902284a6653cc58ad52388db58f4c",
      "inherited_by_current_laro": false
    }
  },
  "swe_bench": {
    "laro_lite": "No valid tier score.",
    "laro_med": "12.5% clean development subset; 20.83% oracle-assisted development subset. Neither is leaderboard-comparable.",
    "laro_max": "Not run.",
    "note": "Development pilots and projections are not leaderboard-comparable and are not marketed as release scores."
  },
  "weight_training": {
    "adapter_win_proven": false,
    "note": "No trained adapter beat the incumbent. No adapter result is marketed as an improvement."
  },
  "runtime_caveat": "Historical MLX and Ollama scores cannot be attributed to the current GGUF release without a reproducible identity-matched run. Simulations are not measured gains.",
  "historical_system_course": {
    "tier": "Laro Lite",
    "checkpoint": "mlx-community/gemma-4-e2b-it-4bit",
    "result": "0/3",
    "score_pct": 0,
    "scope": "latest completed small local production repair course; not SWE-bench or a frontier benchmark",
    "date": "2026-08-02",
    "temporal_note": "This run predates the assertion-guided repair lane. A later bounded candidate tournament passed one narrow fixture, but the complete three-task course has not been rerun."
  },
  "internal_project_checks": {
    "date": "2026-08-09",
    "results": [
      "40/41",
      "38/41",
      "16/16 CRM DOM checks",
      "15/15 one snake-game app/harness comparison"
    ],
    "limitations": [
      "Several project runs reached the 1,500-second limit.",
      "DOM checks do not prove production authentication or tenant isolation.",
      "One artifact's parity does not establish parity for all tasks."
    ]
  }
}